Determining generic items for orders on an online concierge system
An online system provides options for selection by a user. The online system receives a query entered on a client device. The online system queries an item database to retrieve a set of items related to the query and assigns each item to a product category in a predefined taxonomy that maps items to product categories. The online system inputs each item into a prediction model trained to predict a probability that an item is available at a warehouse location. The online system determines that a first product category has low availability based on predicted probabilities for items in the first product category. Responsive to determining that a first product category has low availability, the online system generates a generic item for the first product category and sends a list of items including the generic item to the client device for display responsive to the query.
1. A method comprising, at a computer system comprising a processor and memory:
receiving a query sent from a user device to an online concierge system, wherein the query comprising a string of text entered by a user to the user device for searching the online concierge system;
identifying a set of items matching the query, wherein each item of the set of items corresponds to a product category of a plurality of product categories;
generating an availability prediction for each item of the set of items by applying a prediction model to each item of the set of items, wherein the availability prediction for each item of the set of items is a prediction of the likelihood that the item is available at a third-party warehouse location, and wherein the prediction model is a machine learning model trained to predict a probability that an item is available at a third-party warehouse location based on the third-party warehouse location and a plurality of characteristics associated with the item, wherein training the machine learning model comprises:
obtaining a plurality of training examples, wherein each of the plurality of training examples comprises training data describing an availability of an item included in a previous order,
for each training example of the plurality of training examples, updating the prediction model by:
inputting the training data for the training example into the prediction model, wherein the prediction model generates a probability indicative of whether the item is available at a third-party warehouse location;
computing a score for the training example based on the generated probability and the availability in the training data of the training example; and
updating the prediction model based on the computed score, and
repeating the updating until a confidence score associated with the probability that an item is available is above a threshold;
determining that the product category has low availability based on the generated availability predictions for the set of items;
responsive to determining that the product category has low availability, generating a generic item for the product category; and
sending a list of items to the user device for display to the user, wherein the list of items includes the generic item as one of the items.
2. The method of claim 1 , further comprising:
storing a hierarchical taxonomy of product categories, wherein the hierarchical taxonomy maps items to product categories in the plurality of product categories.
3. The method of claim 2 , wherein the product category for the set of items is a leaf category of the hierarchical taxonomy.
4. The method of claim 1 , wherein determining that the product category has low availability comprises:
computing a total predicted likelihood for the product category based on the generated availability predictions for the set of items; and
comparing the total predicted likelihood to a threshold value.
5. The method of claim 1 , further comprising:
accessing a price for each item in the set of items;
computing an interior price range for the product category based on the accessed prices for the set of items; and
sending the interior price range to the user device to be displayed with the generic item.
6. The method of claim 5 , wherein computing the interior price range comprises:
computing a low end of the interior price range by computing a first percentile value of the accessed prices; and
computing a high end of the interior price range by computing a second percentile value for the accessed prices.
7. The method of claim 5 , wherein accessing a price for each item in the set of items comprises:
determining a most common size of the set of items for the price category; and
accessing a price associated with the most common size for each item in the set of items.
8. The method of claim 5 , further comprising:
receiving an indication from the user device that the generic item was added to an order placed by the user;
determining a maximum price for the generic item based on the accessed prices for the set of items; and
transmitting the maximum price to a user device associated with a picker for display to the picker along with the generic item.
9. The method of claim 8 , wherein the maximum price is a high end of the interior price range.
10. A non-transitory computer-readable medium storing instructions that, when executed by a processor, cause the processor to:
receive a query sent from a user device to an online concierge system, wherein the query comprising a string of text entered by a user to the user device for searching the online concierge system;
identify a set of items matching the query, wherein each item of the set of items corresponds to a product category of a plurality of product categories;
generate an availability prediction for each item of the set of items by applying a prediction model to each item of the set of items, wherein the availability prediction for each item of the set of items is a prediction of the likelihood that the item is available at a third-party warehouse location, and wherein the prediction model is a machine learning model trained to predict a probability that an item is available at a third-party warehouse location based on the third-party warehouse location and a plurality of characteristics associated with the item, wherein training the machine learning model comprises:
obtaining a plurality of training examples, wherein each of the plurality of training examples comprises training data describing an availability of an item included in a previous order,
for each training example of the plurality of training examples, updating the prediction model by:
inputting the training data for the training example into the prediction model, wherein the prediction model generates a probability indicative of whether the item is available at a third-party warehouse location;
computing a score for the training example based on the generated probability and the availability in the training data of the training example; and
updating the prediction model based on the computed score, and
repeating the updating until a confidence score associated with the probability that an item is available is above a threshold;
determine that the product category has low availability based on the generated availability predictions for the set of items;
responsive to determining that the product category has low availability, generate a generic item for the product category; and
send a list of items to the user device for display to the user, wherein the list of items includes the generic item as one of the items.
11. The computer-readable medium of claim 10 , further storing instructions that cause the processor to: store a hierarchical taxonomy of product categories, wherein the hierarchical taxonomy maps items to product categories in the plurality of product categories.
12. The computer-readable medium of claim 11 , wherein the product category for the set of items is a leaf category of the hierarchical taxonomy.
13. The computer-readable medium of claim 10 , wherein the instructions for determining that the product category has low availability further cause the processor to:
compute a total predicted likelihood for the product category based on the generated availability predictions for the set of items; and
compare the total predicted likelihood to a threshold value.
14. The computer-readable medium of claim 10 , further storing instructions that cause the processor to:
access a price for each item in the set of items;
compute an interior price range for the product category based on the accessed prices for the set of items; and
send the interior price range to the user device to be displayed with the generic item.
15. The computer-readable medium of claim 14 , wherein the instructions for computing the interior price range further cause the processor to:
compute a low end of the interior price range by computing a first percentile value of the accessed prices; and
compute a high end of the interior price range by computing a second percentile value for the accessed prices.
16. The computer-readable medium of claim 14 , wherein the instructions for accessing a price for each item in the set of items further cause the processor to:
determine a most common size of the set of items for the price category; and
access a price associated with the most common size for each item in the set of items.
17. The computer-readable medium of claim 14 , further storing instructions that cause the processor to:
receive an indication from the user device that the generic item was added to an order placed by the user;
determine a maximum price for the generic item based on the accessed prices for the set of items; and
transmit the maximum price to a user device associated with a picker for display to the picker along with the generic item.
18. A system comprising:
a processor; and
a non-transitory computer-readable medium storing instructions that, when executed by the processor, cause the processor to:
receive a query sent from a user device to an online concierge system, wherein the query comprising a string of text entered by a user to the user device for searching the online concierge system;
identify a set of items matching the query, wherein each item of the set of items corresponds to a product category of a plurality of product categories;
generate an availability prediction for each item of the set of items by applying a prediction model to each item of the set of items, wherein the availability prediction for each item of the set of items is a prediction of the likelihood that the item is available at a third-party warehouse location, and wherein the prediction model is a machine learning model trained to predict a probability that an item is available at a third-party warehouse location based on the third-party warehouse location and a plurality of characteristics associated with the item, wherein training the machine learning model comprises:
obtaining a plurality of training examples, wherein each of the plurality of training examples comprises training data describing an availability of an item included in a previous order,
for each training example of the plurality of training examples, updating the prediction model by:
inputting the training data for the training example into the prediction model, wherein the prediction model generates a probability indicative of whether the item is available at a third-party warehouse location;
computing a score for the training example based on the generated probability and the availability in the training data of the training example; and
updating the prediction model based on the computed score, and
repeating the updating until a confidence score associated with the probability that an item is available is above a threshold;
determine that the product category has low availability based on the generated availability predictions for the set of items;
responsive to determining that the product category has low availability, generate a generic item for the product category; and
send a list of items to the user device for display to the user, wherein the list of items includes the generic item as one of the items.